We built Frank as an experiment in October last year. A mortgage broker in Sydney wanted an AI agent that could handle SMS threads the same way a human BDM would: chase cold leads, answer rate questions at 9pm, book discovery calls without the back-and-forth dance. Three months later, Frank had sent 2,187 messages and booked 114 qualified appointments. Forty-seven of those turned into settled loans.

What surprised us wasn't the volume. It was the conversion pattern. SMS-to-call worked when the agent knew when to stop typing and ring the lead. This is what we learned from 90 days of real broker conversations.

The brief: a broker who hated phone tag

The operator runs a small Sydney brokerage. Two loan writers, one admin, about 60 settlements a year. His pain point was simple: half his inbound leads came from comparison sites at odd hours. By the time someone rang back the next morning, the lead had already spoken to three other brokers or lost interest.

He wanted an agent that could respond to an 11pm SMS inquiry with something better than "we'll call you tomorrow". Not a chatbot. A system that sounded human, asked qualifying questions, and escalated to a live broker only when the lead was warm.

We configured Frank with a hybrid ruleset: start every conversation over SMS, probe for loan size and timeframe in two or three exchanges, then offer a phone appointment if the lead looked real. If the lead asked a complex question mid-thread, Frank would say "Let me ring you now, easier to explain" and dial within 90 seconds.

What the first 30 days looked like

Frank sent 683 messages in October. Most were replies to cold web leads, a few were follow-ups to stale CRM contacts the broker had flagged. The early pattern was predictable:

• Lead texts "what's your best rate for investment property"
• Frank replies in under a minute, asks loan amount and whether they've got pre-approval elsewhere
• Lead either ghosts or answers
• If they answer, Frank books a 15-minute Calendly slot or offers to ring immediately

Twenty-eight calls were booked that month. Nineteen showed up. The broker closed six loans from that cohort by early December.

The surprise was how often leads preferred the SMS opener. One message thread we reviewed started at 10.40pm on a weeknight. The lead had filled a form on a comparison site, probably from bed. Frank asked three questions over SMS, confirmed the lead wanted to refinance a $680k mortgage, then said "I can call you tomorrow at 9am or now if you're still up". The lead chose 9am. That loan settled six weeks later.

SMS-to-call handoff: the pattern that worked

We expected Frank to book most appointments via text and leave the actual call for later. Instead, we saw a different behaviour emerge. About 40% of qualified leads said yes when Frank offered to "ring you now, it's quicker than typing".

The trick was timing. Frank only made the offer after the lead had answered at least two questions and shown intent. If someone replied "just browsing", Frank stayed in SMS mode and sent a soft follow-up two days later. If they replied with a dollar figure or a settlement deadline, Frank pivoted to voice within one or two exchanges.

This hybrid approach converted better than pure SMS or pure outbound cold-calling. The lead had already demonstrated they were paying attention. The phone call felt like a natural escalation, not an interruption.

The numbers after 90 days

By the end of December, Frank had handled 2,187 SMS exchanges and placed 114 qualifying calls. Sixty-eight of those calls turned into live broker handoffs. Forty-seven deals settled by mid-February, worth about $28 million in total loan value.

The maths works because the broker only paid for successful outcomes. At VoxReach pricing, SMS costs $0.60 per message and outbound mobile calls run $1.32 per minute. Frank's total cost for the quarter was under $4,800. The brokerage earned roughly $140k in trail and upfront commissions from those 47 settlements.

What caught the operator off-guard was how few leads complained about talking to an AI. We tracked sentiment in every thread. Three people asked "is this a bot" in the first month. Frank answered honestly: "Yep, I'm an AI assistant for the team. Still happy to help or I can pass you straight to a human." All three stayed in the conversation.

What surprised us

Two things. First, the 2-way SMS model worked better for mortgage broking than for trades or retail, probably because loan inquiries happen outside business hours and the decision cycle is longer. A lead who texts at 11pm doesn't need an immediate quote; they need proof someone is listening.

Second, the handoff moment mattered more than the script. Frank's voice is calm, Australian, never rushed. When a lead heard the same tone over the phone that they'd seen in SMS, trust carried over. The call we listened to last Tuesday started with the lead saying "oh good, you sound normal". That's the bar.

How to set this up for your business

If you run a brokerage, agency, or consulting practice where leads trickle in at odd hours, the hybrid SMS-to-call model is worth testing. You need a CRM that logs message threads, a calendar tool for booking, and an AI agent that knows when to stop typing and dial.

VoxReach handles all three. Setup fee is $5,500. Inbound calls from $0.42/min, outbound to AU mobiles $1.32/min, SMS $0.60/msg. Native Australian voice, 30-day account with.

Sign up at app.voxreach.com.au/signup and configure your first hybrid agent in under an hour. Or ring +61 2 5926 2202 to talk to Frank himself. Same platform, same voice, live demo.